Factors Determining Quality of Red Wine
# load data
rw <- read.csv('wineQualityReds.csv')We will be examining a dataset (provided by Cortez et al) consisting of 1599 observations of Portugese red wine. Each observation contains information about its physicochemical properties and a quality rating (0 to 10) calculated from the median of at least three expert evaluations.
Univariate Exploration
Let’s examine the columns, their data types, and values.
str(rw)## 'data.frame': 1599 obs. of 13 variables:
## $ X : int 1 2 3 4 5 6 7 8 9 10 ...
## $ fixed.acidity : num 7.4 7.8 7.8 11.2 7.4 7.4 7.9 7.3 7.8 7.5 ...
## $ volatile.acidity : num 0.7 0.88 0.76 0.28 0.7 0.66 0.6 0.65 0.58 0.5 ...
## $ citric.acid : num 0 0 0.04 0.56 0 0 0.06 0 0.02 0.36 ...
## $ residual.sugar : num 1.9 2.6 2.3 1.9 1.9 1.8 1.6 1.2 2 6.1 ...
## $ chlorides : num 0.076 0.098 0.092 0.075 0.076 0.075 0.069 0.065 0.073 0.071 ...
## $ free.sulfur.dioxide : num 11 25 15 17 11 13 15 15 9 17 ...
## $ total.sulfur.dioxide: num 34 67 54 60 34 40 59 21 18 102 ...
## $ density : num 0.998 0.997 0.997 0.998 0.998 ...
## $ pH : num 3.51 3.2 3.26 3.16 3.51 3.51 3.3 3.39 3.36 3.35 ...
## $ sulphates : num 0.56 0.68 0.65 0.58 0.56 0.56 0.46 0.47 0.57 0.8 ...
## $ alcohol : num 9.4 9.8 9.8 9.8 9.4 9.4 9.4 10 9.5 10.5 ...
## $ quality : int 5 5 5 6 5 5 5 7 7 5 ...We see 11 continuous variables pertaining to physicochemical properties and it would be excessive to create histograms and calculate descriptive statistics for all of them. Fortunately the dataset builders have identified specific variables that are expected to affect the perceived quality of wine. We will also examine secondary properties to see if they too bear a relationship to perceived quality.
Begin with volatile acidity measured in grams per cubed decimetre, Cortez et al claim that too high levels ‘can lead to an unpleasant vinegar taste’.
qplot(x = volatile.acidity, data = rw, binwidth = 0.1) +
scale_x_continuous(breaks = seq(0, 1.5, 0.2))
summary(rw$volatile.acidity)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 0.1200 0.3900 0.5200 0.5278 0.6400 1.5800So the distribution clusters around the mean, with the graph skewing slightly right. Let’s examine citric acid, which Cortez et al claim adds freshness and flavour to wines when in small quantities.
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 0.000 0.090 0.260 0.271 0.420 1.000This distribution does not have a strong central tendency and skews to the right. Let’s look at residual sugar, also measured in grams per cubic decimetre: Cortez et al claim wines rarely have less than 1g/litre or more than 45g/litre which is considered too sweet
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 0.900 1.900 2.200 2.539 2.600 15.500This is indeed the case as the authors claim, our lowest observation value is only 0.9 grams and the highest is 15.5 grams. Now, let’s examine chlorides, which is the amount of salt in the wine. Obviously wine wouldn’t taste too good if it was salty.
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 0.01200 0.07000 0.07900 0.08747 0.09000 0.61100Again, the distribution is similar to residual sugars. Let’s consider free sulfur dioxide, which Cortez et al claim prevents microbial growth and oxidation of wine and is measured in parts per million (ppm). Once it crosses 50 ppm, the taste and smell of wine changes.
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 1.00 7.00 14.00 15.87 21.00 72.00There are very few observations crossing 50 ppm. Now let’s examine the pH or how acidic or basic the wine is: a pH of 0 is highly acidic whilst that of 14 is very basic.
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 2.740 3.210 3.310 3.311 3.400 4.010Our wines are quite acidic, most of them cluster around a pH of 3 to 4, without any exceeding that range. Finally, we consider the alcohol content of our wines, with values represented as percentages.
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 8.40 9.50 10.20 10.42 11.10 14.90So our wines have higher alcohol content than most beers, with a mean of 10.42% and the minimum being 8.4%, there is even a wine with an alcohol content of 14.9%! I would certainly like to taste it, later we shall see if higher alcohol content increases quality ratings.
Finally, lets look at the distribution of quality ratings for our Portugese wines. Note this is a modified quality rating which is calculated from taking the median of at least three reviews given by wine experts.
ggplot(rw, aes(x=factor(quality))) + geom_bar()
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 3.000 5.000 6.000 5.636 6.000 8.000So the vast majority of our wines are rated between 5 or 6, which is fair. The lowest rating is a 3, but there are no perfect ratings either, with the highest being an 8. This could be due to the prudent judgements given by experts, and also because the final rating is calculated from the median.
Let’s create a categorical variable called ‘class’ based on ratings given by experts. Our univariate plot shows ratings for all wines falling between 3 - 8, so a class of ‘poor’ can be assigned for ratings from 3 - 4, ‘fair’ for 5 - 6, and ‘good’ for 7 - 8. We can then use this to see if the strength of the relationship between physicochemical properties differs according to wine class.
rw$class <-
ifelse(rw$quality > 6, 'Good', ifelse(rw$quality > 4, 'Fair', 'Poor'))Now that we have created this new categorical variable, let’s plot out the distribution
Univariate Analysis
Dataset Structure
Our dataset consists of 1599 observations with 11 variables, all represented by values that are continuous and numerical. There are no categorical variables, and all are in consistent units without any missing values. All variables relate to either the physicochemical properties of the wine, or the quality rating.
Features of Interest
Most wines are acidic (low pH), have low residual sugar (below the 45g threshold that would be considered too sweet), and low free sulfur dioxide (less than the 50ppm that would be noticeable in taste and smell). On average the alcohol content is around 10%. Most wines have a quality rating of 5 - 6, which we define as ‘fair’. A natural avenue for exploration in this data set is how these physicochemical properties affect the quality rating. But we could also consider how such properties are related to each other, and whether such relationships are affected by wine quality.
Variables
A new categorical variable was created based on the quality rating the wine received, this will allow us to see if the strength of relationships between physicochemical properties varies by wine quality. No data wrangling was necessary because the variables are already in consistent units without missing values. Although some variable distributions like residual sugars and chlorides had outliers or a long tail, this didn’t prevent us from clearly observing the values distribution thus no transformations were necessary.
Bivariate Exploration
Cortez et al have already described the physicochemical properties they expect to influence quality ratings, and we have examined their distributions using univariate plots. Here we will use a scatter plot matrix to gain a high level overview of the relationship between these primary physicochemical properties and quality ratings, in addition to how they relate to each other.
primary = c("free.sulfur.dioxide", "volatile.acidity", "citric.acid",
"residual.sugar", "alcohol", "quality")
ggpairs(rw[primary])
Let us get to the key question of how primary physicochemcial properties affect quality ratings. It can be seen that alcohol percentage is strongly correlated with quality (0.476).
ggplot(aes(alcohol, quality), data = rw) +
geom_jitter(alpha = 0.3) 
Here is visual evidence that wines with higher alcohol percentages attract higher quality ratings. This is one aspect that Cortez et al have not remarked upon. Residual Sugar on the other hand is shown to be very weakly correlated with quality ratings, at only 0.0137- this is likely because there is very little variation in residual sugar values and they never exceed the threshold (45g) at which the authors state would degrade the wine tasting experience.
Citric acid is also found to be positively correlated with quality ratings, at 0.226.
Higher level of citric acid are associated with higher quality ratings and this is consistent with the Cortez et al’s comment that the presence of citric acid adds freshness to the wine.
On the other hand, volatile acidity is negatively correlated with quality ratings, at -0.391.
This is consistent with Cortez et al’s observation that higher levels of such acetic acid can lead to an unpleasant and sour taste. Our matrix also presents a strong negative correlation between citric acid and volatile acidity at -0.552
The more citric acid, the less volatile acidity. Further on we will attempt to see if this relationship is particularly strong for higher rated wines, is there some kind of special production process that gives certain wines an advantage?
Now that we have examined the primary physicochemical properties, we must consider secondary physicochemical properties that Cortez et al were less attentive towards. Do they have any relationship with quality ratings? Do they have any relationship with each other?
The matrix show a positive relationship between sulphates and quality ratings, with a correlation of 0.251.
It is not clear why adding sulphates improves quality ratings, Cortez et al mention that doing so leads to an increase in free sulfur dioxide which prevents microbial growth and oxidation of wine, but previous calculations show that the relationship between free sulfur dioxide and rated quality is weakly negative.
The matrix also shows a strong correlation between density and fixed.acidity at 0.668.
This show wine density increasing with higher levels of non-evaporative acidity. We will see if the strength of this relationship changes with the quality rating of the wine.
Note that a highly negative relationship exists between pH and fixed acidity with a correlation of -0.683, but it is self explanatory why this is so: a lower pH is by definition more acidic.
Bivariate Analysis
The dataset authors already commented on the primary physicochemical properties they believed influenced quality ratings. Our bivariate plots and calculations largely confirmed their analysis: for example volatile acidity (or acetic acid) has a -0.391 correlation with quality ratings, citric acid is positively correlated with quality ratings at 0.226. However one relationship Cortez et al didn’t discuss was that between alcohol and quality rating, where there was a strong positive correlation of 0.476.
Citric acid has a strong negative correlation with volatile acidity: -0.552. Any wine that can leverage this relationship can substantially improve its ratings from experts. We will be doing further investigation to see if higher quality wines embody this relationship. Density is highly correlated with fixed acidity (0.668), further investigation will be conducted to see if the strength of the relationship varies with wine quality.
Multivariate Plots Section
Based on our bivariate plots, we know that there is a strong negative correlation between citric acid and volatile acidity. Does the magnitude of this negative relationship differ according to the quality rating of the wine?
ggplot(aes(x = citric.acid, y = volatile.acidity), data = rw) +
geom_point(aes(color = class), alpha = 0.5)
Our plot demonstrates that highly rated wines (represented in green) tend to have high citric acid and low volatile acidity, whereas poorly rated wines (represented in blue) tend to have low citric acid and high acetic acid, whereas fairly rated wines are in between.
We also found a strong correlation between density and fixed acidity: 0.668. Does the strength of the relationship vary depending on the class of wine?
Our plot demonstrates that the strength of the relationship between density and fixed acidity is not contingent upon the class of wine.
Multivariate Analysis
Our previous bivariate plots showed that citric acid was positively correlated with quality ratings, and volatile acidity was negatively correlated with quality ratings. The plots also showed that volatile acidity and citric acid have a strongly negative relationship with each other. The multivariate plot shows that this negative correlation is particularly pronounced with higher quality wines.
The fact that ‘good’ class wines, or higher quality wines embody a particularly strong negative relationship between citric acid and acetic acid, seems to suggest that this is one of the key mechanisms by which they attract high ratings: by lowering the amount of acetic acid (which Cortez et al claim lead to a sour taste) and raising the amount of citric acid (which is claimed to lead to fresh taste). There may, in fact, be some kind of special production technique at play.
Final Plots and Summary
Plot One
qplot(x = factor(quality), data = rw, main="Distribution of Quality Ratings for Wine",
ylab="Count", xlab="Quality Rating (0 to 10)")
The plot illustrates that ratings for our 1599 wine observations are surprisingly stable, the vast majority given are 5 or 6. The lowest rating is a 3 and the highest is an 8. This may be due to calculation of ratings based on the median of three, rather than as a mean. It may also say something about the nature of reviewers or the nature of the wine. We need more information about how the data was collected to make further inferences.
Plot Two
ggplot(aes(factor(quality), alcohol), data = rw) + geom_jitter(alpha = 0.4) +
geom_boxplot(aes(group = quality), alpha = 0.5, color = 'blue') +
ggtitle("Alcohol Content per Wine Quality") +
ylab("Alcohol (%)") +
xlab("Quality Rating (0 to 10)")
A key finding from our study, and one not mentioned by the authors, is the positive relationship between alcohol and quality ratings (0.476). The higher the quality rating, the higher the alcohol content tends to be.
Plot Three
ggplot(aes(x = citric.acid, y = volatile.acidity), data = rw) +
geom_point(aes(color = class), alpha = 0.6) + xlab("Citric Acid (g/cubic decimetre)") +
ylab("Volatile Acidity (g/cubic decimetre)") +
ggtitle("Relationship between Citric Acid and Volatile Acidity")
In general citric acid is negatively correlated with volatile acidity (or acetic acid): -0.552. Higher levels of citric acid are positively associated with quality ratings, the opposite is true for volatile acidity. The nature of the relationship is particularly strong for high quality wines, which tend to have the highest amounts of citric acid, and the lowest amount of acetic acid. Have those producers found some special process for making wines that win the praise of critics?
Reflection
Exploration of the dataset has had benefits and challenges. The easy part has been the lack of missing values and cleanliness and consistency in the format of the data. The challenge has been lack of categorical variables that could provide further insight. For example, knowing the region or the type of soil in which the wines were grown could have yielded additional insights in terms of understanding relationships involving quality ratings or physicochemical properties. Nevertheless we have been able to gain some insights about what drives wines to gain higher quality ratings: namely alcohol content, and the inverse relationship between citric and acetic acid. In the future, taking the mean rather than the median of expert ratings may provide a more accurate picture of how the wines were received. Adding additional categorical information such as fermentation method, soil type, producer, would help us find additional insights into what makes great wine, and what produces certain physicochemical properties.